Moodle Mcp Server Aql

moodle-mcp-server — AI-Powered LMS Analytics

Ask your Moodle instance anything. Get structured answers in seconds.

moodle-mcp-server is an open-source MCP (Model Context Protocol) server that connects AI agents directly to Moodle's Web Services API. It is the first reference implementation of the Agentic Query Layer (AQL) pattern — a software design pattern for giving large language models safe, permission-aware, read-only access to structured data.

Instead of learning report builders, writing SQL, or exporting CSVs, you ask questions in plain English — the AI agent queries your LMS and returns structured data.

AQL pattern: https://csmediapro.com/notes/agentic-query-layer Project home: https://csmediapro.com/products/moodle-mcp-server npm package: https://www.npmjs.com/package/moodle-mcp-server-aql MCP Registry: io.github.csmediapro/moodle-mcp-server-aql

Features

• Core query tools — course catalog, user enrollment, assignments, category navigation, site metadata, user lookup, and cache management
• Premium reporting plugins — optional tools such as course completion reports and recent activity can be attached through the plugin runtime
• LLM-agnostic — works with Claude, GPT, Gemini, Ollama, or any MCP-compatible AI client
• Zero LMS modification — uses Moodle's existing Web Services API, no plugin installation required
• Compliance-ready data handling — secure, efficient data processing with privacy by design
• Read-only — never modifies Moodle data, safe for production
• Plugin-extensible — drop new tool modules into a directory; tools and optional agent routing hints auto-register at runtime

Quick Start

Prerequisites

• Node.js 20+
• A Moodle instance with Web Services enabled
• A Moodle API token (Site administration → Plugins → Web services → Manage tokens)

Use With An MCP Client

Most users launch the server through an MCP client such as Claude Desktop:

{  "mcpServers": {    "moodle-mcp-server-aql": {      "command": "npx",      "args": ["-y", "moodle-mcp-server-aql"],      "env": {        "MOODLE_URL": "https://your-moodle-instance.example",        "MOODLE_TOKEN": "your-moodle-web-services-token"      }    }  }}

Setup From Source

# Clone the repogit clone https://github.com/csmediapro/moodle-mcp-servercd moodle-mcp-server
# Install dependenciesnpm install
# Configurecp packages/server/.env.example packages/server/.env# Edit .env: add your MOODLE_URL and MOODLE_TOKEN
# Run (stdio mode)npm run server:buildnode packages/server/dist/index.js

User field schema

User field display settings are generated per Moodle instance and stored locally at packages/server/data/user-field-schema.json. This file is intentionally ignored by git because it can include site-specific custom profile fields.

After connecting to a Moodle site, run the refresh_user_field_schema tool once to discover available standard and custom user fields. A minimal example shape is included at packages/server/data/user-field-schema.example.json.

Separately installed plugins can use this schema to flatten custom profile fields into stable keys such as school. The premium user-directory plugin, for example, uses it to cache and summarize full-user directory fields outside the OSS core.

Config identity

The core owns two different server identity fields:

• server.id — stable machine identity, for example mcp_8f3k2q9x
• server.name — human-facing display label

If server.id is missing, the core generates one once and persists it to the resolved config file before startup continues.

Environment overrides:

• MOODLE_MCP_CONFIG or MOODLE_MCP_SERVER_CONFIG — choose the config file path
• SERVER_ID — explicit server.id override
• SERVER_NAME — explicit server.name override
• SERVER_VERSION — explicit server.version override

If server.id is missing and the resolved config path is not writable, startup fails deliberately.

Using the reference client

# From the project rootcp packages/client/.env.example packages/client/.envnpm run client:dev# Open http://localhost:3000

The client will auto-detect your Moodle instance and present a chat interface where you can ask questions in plain English.

Connecting an LLM

moodle-mcp-server needs an AI model to power the natural-language interface. You bring the model — the moodle-mcp-server core and reference client support any MCP-compatible provider.

Option 1: Run Locally (Recommended for Speed & Privacy)

Running a local model keeps all data on your own hardware — nothing leaves your network. Modern quantized models run well on consumer GPUs and even CPU-only setups.

Performance: A quantized 24B model on a single RTX 3090 delivers ~1.5-second responses after the first query — faster than most cloud APIs once the system is initialized.

Via Ollama (easiest)

# Install Ollama: https://ollama.comollama pull gemma3:12b      # Fast, reliable tool use (~200ms TTFT)ollama pull qwen3:14b       # Strong reasoning, good for complex queriesollama pull deepseek-r1:14b # Excellent at multi-step chains

Then point the reference client at http://localhost:11434 (Ollama's default).

Via llama.cpp (maximum control)

# Download a GGUF model (example: Devstral 24B Q4)# Run the llama.cpp server:llama-server -m devstral-24b-Q4_K_M.gguf --ctx-size 60000 --port 8080

Point the reference client at http://localhost:8080/v1.

Recommended local models

Option 2: Cloud Providers

Anthropic (Claude):

export ANTHROPIC_API_KEY=sk-ant-...

Select "Anthropic" in the reference client's provider dropdown. Claude Sonnet offers the most reliable tool-calling behavior.

OpenAI (GPT):

export OPENAI_API_KEY=sk-...

Select "OpenAI" in the provider dropdown. GPT-4o performs well on structured queries.

Ollama Cloud: Uses the same API as local Ollama, hosted at https://ollama.com/v1. Good middle ground — faster than local cold starts, more private than big cloud providers.

Option 3: Claude Desktop (Direct MCP)

Claude Desktop connects to the moodle-mcp-server core directly over stdio — no reference client needed.

Add to your Claude Desktop config (claude_desktop_config.json):

{  "mcpServers": {    "moodle-mcp-server": {      "command": "node",      "args": ["/path/to/moodle-mcp-server/packages/server/dist/index.js"],      "env": {        "MOODLE_URL": "https://your-moodle-instance.com",        "MOODLE_TOKEN": "your-api-token"      }    }  }}

Restart Claude Desktop. The server's tools will appear in Claude's tool list — ask questions directly.

Tools

Core Tools (free, open source — AGPL)

Premium Plugins (available separately)

• Advanced Reporting — gradebooks, cross-course comparison, custom report builder
• User Analytics — progress tracking, engagement scoring, risk flags
• User Directory — cached directory listing and structured filtering across standard and custom profile fields
• Compliance Pack — certification tracking, expiration alerts, audit exports

Architecture

This project is a reference implementation of the Agentic Query Layer (AQL) pattern — a layer that sits between an AI agent and your data, providing permission-aware, read-only, plain-language access without exposing the underlying data to the model.

User (plain English question)    │    ▼AI Agent (Claude / GPT / Gemini / Ollama / local)    │    ▼  MCP Protocol`moodle-mcp-server`  ← AQL connector for Moodle    ├── Tool Registry (core + plugins)    ├── Agent Runtime Config (core + plugin rules)    ├── Optimized Data Layer (secure, efficient data handling)    └── Moodle Client (REST API calls)    │    ▼Moodle Web Services API

Transport modes

• Stdio — dist/index.js — runs as a subprocess, used by Claude Desktop and similar clients

The OSS core intentionally ships with stdio only. Any network-facing wrapper, remote supervision, or premium plugin attachment belongs in a separate commercial node agent or wrapper.

Plugin docs

• Plugin Contract
• Creating Plugins
• Hello Plugin Example

License

AGPL v3 — see LICENSE.

This means you can:

• ✅ Use the moodle-mcp-server core for free, in any environment
• ✅ Modify the source code for your needs
• ✅ Build and distribute derivative works

You cannot:

• ❌ Repackage the moodle-mcp-server core as a closed-source competing commercial product
• ❌ Offer it as a network service without sharing your modifications

Trademark Notice

Moodle is a trademark of Moodle Pty Ltd. moodle-mcp-server is an independent CSMediaPro project and is not affiliated with, endorsed by, sponsored by, or officially connected to Moodle Pty Ltd or the Moodle project. The name is used descriptively to identify compatibility with Moodle LMS.

Built by CSMediaPro

moodle-mcp-server is built and maintained by CSMediaPro, a software development company specializing in AI integration, systems engineering, and workflow automation. It is the first reference implementation of the Agentic Query Layer (AQL) pattern.

• AQL pattern: https://csmediapro.com/notes/agentic-query-layer
• Contact: [email protected]